Skip to main content

Python Client Library

Python implementation of the agent-backend package. See the main README for an overview, quick start, and core usage.

Package Info

Field Value
Package agent-backend
Registry PyPI
Manager uv / pip
Test runner pytest
Build uv build
Linter ruff
Type checker ty
Source packages/agent-backend/python/agent_backend/
Tests packages/agent-backend/python/tests/

Advanced Features

Environment Variables

Scoped backends support custom environment variables that apply to all commands:

from agent_backend import ScopeConfig

scoped_backend = backend.scope("projects/my-app", ScopeConfig(
    env={
        "PYTHONPATH": "/workspace/lib",
        "API_KEY": "secret",
        "DATABASE_URL": "postgres://...",
    }
))

await scoped_backend.exec("python -m build")  # uses custom env

Operations Logging

from agent_backend import ConsoleOperationsLogger, ScopeConfig

scoped_backend = backend.scope("project", ScopeConfig(
    operations_logger=ConsoleOperationsLogger()
))

await scoped_backend.exec("pip install -r requirements.txt")
# Logs: [AgentBackend] exec: pip install -r requirements.txt

Binary Data

from agent_backend import ReadOptions

image_data = await backend.read("logo.png", ReadOptions(encoding="buffer"))
tarball = await backend.exec("tar -czf - .", ExecOptions(encoding="buffer"))

Timeouts

from agent_backend import RemoteFilesystemBackend, RemoteFilesystemBackendConfig

backend = RemoteFilesystemBackend(RemoteFilesystemBackendConfig(
    root_dir="/tmp/agentbe-workspace",
    host="server.com",
    auth_token="...",
    operation_timeout_ms=300_000,  # 5 minutes
    max_output_length=10 * 1024 * 1024,  # 10MB
))

Backend Connection Pooling

See docs/connection-pooling.md for BackendPoolManager usage, key-based pooling, idle cleanup, and graceful shutdown.

Examples

Code Execution Sandbox

from agent_backend import LocalFilesystemBackend, LocalFilesystemBackendConfig, IsolationMode

sandbox = LocalFilesystemBackend(LocalFilesystemBackendConfig(
    root_dir="/tmp/agentbe-workspace",
    isolation=IsolationMode.AUTO,
))

user_code_backend = sandbox.scope(f"users/{user_id}")
await user_code_backend.write("script.py", untrusted_code)
result = await user_code_backend.exec("python script.py")

Multi-tenant SaaS

from agent_backend import RemoteFilesystemBackend, RemoteFilesystemBackendConfig

# Separate backend per organization
org1_backend = RemoteFilesystemBackend(RemoteFilesystemBackendConfig(
    root_dir="/var/saas/org1",
    host="org1-server.example.com",
    auth_token="...",
))

org2_backend = RemoteFilesystemBackend(RemoteFilesystemBackendConfig(
    root_dir="/var/saas/org2",
    host="org2-server.example.com",
    auth_token="...",
))

# Scoped backends per user within each org
org1_user1 = org1_backend.scope("users/user1")
org1_user2 = org1_backend.scope("users/user2")

Agent State Management

from agent_backend import MemoryBackend

state = MemoryBackend()

await state.write("agents/agent1/current-task", "building")
await state.write("agents/agent1/progress", "50%")

all_agents = await state.list_keys("agents/")

Error Handling

from agent_backend import BackendError, DangerousOperationError, PathEscapeError

try:
    await backend.exec("rm -rf /")
except DangerousOperationError as e:
    # Command blocked by safety validation
    print("Blocked:", e.operation)
except PathEscapeError:
    # Path attempted to escape scope
    pass
except BackendError as e:
    # General backend error (check e.code)
    print("Error:", e.code, str(e))

Development

Commands

All commands can be run from the monorepo root via Make or from the packages/agent-backend/python/ directory via uv.

Task Make (root) uv (packages/agent-backend/python/)
Build make build-python uv build
Test make test-python uv run pytest
Test (cov) -- uv run pytest --cov=agent_backend --cov-fail-under=80
Lint make lint-python uv run ruff check .
Lint (fix) make lint-fix uv run ruff check --fix .
Typecheck make typecheck-python uv run ty check

Code Style

  • ruff enforces formatting (line-length = 100, target-version = "py311")
  • Type hints on all function signatures -- avoid Any
  • snake_case for functions and variables, PascalCase for classes
  • Dataclasses for all config objects (LocalFilesystemBackendConfig, ScopeConfig, etc.)
  • Custom error classes: BackendError, DangerousOperationError, PathEscapeError
  • Imports sorted with ruff (isort rules enabled)

Testing

Tests live in packages/agent-backend/python/tests/ and use pytest with pytest-asyncio.

asyncio_mode = "auto" is configured in pyproject.toml, so async test functions are detected automatically -- no @pytest.mark.asyncio decorator needed.

Shared fixtures in conftest.py provide pre-configured backends:

@pytest.fixture
def local_backend(tmp_workspace):
    config = LocalFilesystemBackendConfig(
        root_dir=tmp_workspace,
        prevent_dangerous=True,
    )
    return LocalFilesystemBackend(config)

Use unittest.mock.AsyncMock for mocking async methods. Use the shared fixtures (local_backend, memory_backend, tmp_workspace) rather than building backends from scratch.

Running tests:

uv run pytest                            # All tests, single run
uv run pytest -k "safety"               # Filter by pattern
uv run pytest --cov=agent_backend       # With coverage report
uv run pytest -m "not integration"      # Skip integration tests

Gotchas

  • All backend methods are async -- always await them, including read, write, readdir, and exists.
  • MemoryBackend.exec() raises NotImplementedBackendError -- memory backends do not support command execution.
  • Use list_keys(prefix) on MemoryBackend, not list().
  • IsolationMode.AUTO and IsolationMode.NONE are enum members, not string literals.
  • BackendType enum values are "local-filesystem", "remote-filesystem", "memory".
  • Config objects are dataclasses, not dicts -- use keyword arguments (e.g., LocalFilesystemBackendConfig(root_dir=...)).
  • Scoped backends delegate track_closeable() to their parent, so resources are closed when the parent is destroyed.
  • destroy() closes all tracked closeables (MCP clients, transports) before tearing down the backend.
  • Coverage threshold is 80% (--cov-fail-under=80). Remote backend and transport modules are excluded from coverage.
  • ExecOptions and ReadOptions use encoding: Literal["utf8", "buffer"], not Python's standard encoding names.

Release files for agent-backend 0.13.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for agent-backend 0.13.1
File Size Uploaded
agent_backend-0.13.1.tar.gz 34.6 kB Details

Built distribution (wheel)

Table of built distributions (wheels) for agent-backend 0.13.1
File Interpreter ABI Platform
agent_backend-0.13.1-py3-none-any.whl Python 3 none any Details

Total release size: 68.6 kB

Release files / agent_backend-0.13.1.tar.gz

Download URL agent_backend-0.13.1.tar.gz
Size 34.6 kB
Tags Source
SHA-256 checksum
How to use checksums
b1d39827562960955e056cea94b28c7d77e98041a2a747c12d625dae6f51b053
BLAKE2b-256 checksum
How to use checksums
bebbc773829cf3fdc9b05d896b952d34b41e3ef98ef0f5fdf77409f76cfbce48
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 16, 2026.

Transparency log

Release files / agent_backend-0.13.1-py3-none-any.whl

Download URL agent_backend-0.13.1-py3-none-any.whl
Size 34.0 kB
Tags Python 3
SHA-256 checksum
How to use checksums
9744d02affd6da317e01239de6b630fed9064bf7f48abe88a8c3a5a6a074da03
BLAKE2b-256 checksum
How to use checksums
ee8e99bb00e2617bb2635eafe40cf0da05f1512cc769eb5b766ac8775cf2b5a3
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.14

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Sep 16, 2026.

Transparency log

Release history Release notifications | RSS feed

0.13.2

2 release files

This release

0.13.1 This release

2 release files

0.13.0

2 release files

0.12.1

2 release files

0.12.0

2 release files

0.11.2

2 release files

0.11.1

2 release files

0.10.0

2 release files

0.9.8

2 release files

0.9.7

2 release files

0.9.6

2 release files

0.9.5

2 release files

0.9.4

2 release files

0.9.3

2 release files

0.9.2

2 release files

0.9.1

2 release files

0.9.0

2 release files

0.8.9

2 release files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page